Meta launches tool to convert text to speech for more than a thousand languages

2023-05-22 21:06:00

Meta, the company behind Facebook, has expanded its voice-to-text conversion system capable of doing the opposite as well. The new was announced by Meta on its website, where it is said that now the conversion engine developed by the company now supports more than 1,100 languages, which is ten times more than before. There was also an improvement in the language identification model, which is now able to identify 4000 languages.

Meta introduces tool to convert text to speech

Transcription system between texts and voices in several Meta languages ​​is improved and gains support for 1100 languages. Source: Meta

Called Massively Multilingual Speech (MMS), Meta’s project aims to make devices able to understand and produce speech, bringing greater accessibility in various situations where knowledge of a given language is an impediment to communicating or understanding something. Previously, the system developed by the company supported only 100 languages, which represents only a fraction of the more than 7 thousand languages ​​known in the world, but now there has been a huge leap to 1100 languages.

The MMS performs well compared to existing language models for transcription between texts and voices in different languages, according to Meta. Fortunately, the company has announced that it is sharing its templates and their code so that others in the research community can create different software from this project.

Why make a speech model available for multiple languages?

According to Meta, several languages ​​”are in danger of disappearing, and the limitations of current speech recognition and generation technologies will only accelerate this trend”. Through Massively Multilingual Speech (MMS), the company wants to “make it easier for people to access information and allow them to use devices in their preferred language.” According to the company, there are plans to develop “a single model that can solve multiple speech tasks for all languages”.

Today Meta has several separate models for speech recognition, speech synthesis and language identification. However, the company intends to unify all of this in a single model, thus delivering better overall performance.

1684791700
#Meta #launches #tool #convert #text #speech #thousand #languages

Leave a Replay